README.md

metadata

license: other
license_name: gemma-terms-of-use
license_link: https://ai.google.dev/gemma/terms
model-index:
  - name: google-gemma-7b-it-dpo-v1
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 51.54
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=cloudyu/google-gemma-7b-it-dpo-v1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 71.58
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=cloudyu/google-gemma-7b-it-dpo-v1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 53.24
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=cloudyu/google-gemma-7b-it-dpo-v1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 46.85
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=cloudyu/google-gemma-7b-it-dpo-v1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 67.25
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=cloudyu/google-gemma-7b-it-dpo-v1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 27.67
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=cloudyu/google-gemma-7b-it-dpo-v1
          name: Open LLM Leaderboard

this is a DPO fine-tuned model for google/gemma-7b-it using jondurbin/truthy-dpo-v0.1

DPO Trainer
TRL supports the DPO Trainer for training language models from preference data, as described in the paper Direct Preference Optimization: Your Language Model is Secretly a Reward Model by Rafailov et al., 2023. 

 target_modules=[ "gate_proj", "up_proj", "down_proj"]

sample code

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
import math

## v2 models
model_path = "cloudyu/google-gemma-7b-it-dpo-v1"

tokenizer = AutoTokenizer.from_pretrained(model_path, use_default_system_prompt=False)
model = AutoModelForCausalLM.from_pretrained(
        model_path, torch_dtype=torch.bfloat16, device_map='auto',local_files_only=False, load_in_4bit=True
)
print(model)
prompt = input("please input prompt:")
while len(prompt) > 0:
  input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to("cuda")

  generation_output = model.generate(
    input_ids=input_ids, max_new_tokens=500,repetition_penalty=1.2
  )
  print(tokenizer.decode(generation_output[0]))
  prompt = input("please input prompt:")
  

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	53.02
AI2 Reasoning Challenge (25-Shot)	51.54
HellaSwag (10-Shot)	71.58
MMLU (5-Shot)	53.24
TruthfulQA (0-shot)	46.85
Winogrande (5-shot)	67.25
GSM8k (5-shot)	27.67